Differences-in-Differences for staggered policy introduction

[Sandrine Klaus]

Dear Statalist community,

I want to do a Diff-in-diff estimation. My data is on the state-level and compromises the years 1985-2000

State-year-level observations: State:=1,…,16 & year:= 1985,…,2000 (ca. 64’000 firm-year observations)

12 of 16 states implemented a policy. The 12 states that adopted the policy did so (mainly) in different years. The staggering of the policy introduction should allow that firms are first control firms (before the law) and then treatment firms (after the law). Firms located in states that never passed a policy are always control firms.

Non-adopter states have been imputed a “would have adopted” date. This should ensure to define a era (pre-adoption vs post-adoption) variable.

Now, I am not totally sure if my Stata code does account for this staggered policy introduction and, therefore, treats firms before a policy passed as control and thereafter as treatment firms.

Code:   
 gen byte era = (year > adoption_year) // where adoption_year is a variable identifying the actual or would-be adoption of state  
 gen byte group = policy!=0 //where policy indicates whether a state passes a law or not  
 reghdfe outcome i.group##i.era covariates, a(firm year##industry)  vce(cluster state) 

Thank you very much for your support in advance! Kind regards,
Sandrine

[Clyde Schechter]

This code looks to me like it properly accounts for the staggered adoption dates.

I don't use -reghdfe- very often, so I'm not sure that your options there are correctly specified. But you can check that easily enough.

[Sandrine Klaus]

Thank you very much Clyde for your response!

[Jordi Pascual]

Dear all,

I am also trying to do a staggered DiD but I am not complitely sure about how to compute the "would have adopted date". Could someone explain me with an example the way to introduce those dates for the non-adopters?

Thank you!